The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process a "reproduction package" and "PoC" provided by an external caller. This ingestion of untrusted data creates a surface for indirect prompt injection where malicious instructions could be embedded in the PoC or input data to manipulate the agent during the replay process.
Ingestion points: SKILL.md defines the input as a "complete sealed reproduction package" and "PoC".
Boundary markers: The instructions attempt to scope the work by referencing a "sealed" package and specific "project guidance" and "security guidance".
Capability inventory: The agent is instructed to build and execute code, check interface reachability, and write results to a replay.json file.
Sanitization: The skill includes a mandatory step to "Verify the package; read setup and security guidance" and explicitly tells the agent to "Stop execution if prerequisites fail."
[COMMAND_EXECUTION]: The core functionality of the skill requires the agent to "independently build and run" code from the provided PoC. This involves executing arbitrary commands within the environment to verify behavior differences between versions (S0/S1). While this is the intended use case, it remains a high-capability execution path that depends on the integrity of the supplied reproduction package.

replay-regression